Enabling Viewpoint Learning through Dynamic Label Generation

Tensorflow implementation of Enabling Viewpoint Learning through Dynamic Label Generation, published at Eurographics 2021.

M. Schelling, P. Hermosilla, P.-P. Vázquez and T. Ropinski

Paper Pre-Print

Project Page

Prerequisites

• Download and compile the MCCNN library and place it it the MCCNN/ folder.
• Download the data and place it in the viewpoint_learning/data/ folder

This Implementation is in TensorFlow 1 and was tested using TF 1.11 and Python 2.7. For the viewpoint computation methods the OpenGL package for python is required. For training this is not necessary. We recommend to run training in a tf=1.11_gpu docker container.

Example Training

The root directory contains scripts to train viewpoint prediction networks using dynamic label generation with Multiple Labels (ML), Gaussian Labels (GL) and a two staged learning using both (ML-GL).

Reference implementation are given for Single Label (SL), Spherical Regression (SR), Deep Label Distribution Learning (DLDL).

Note: By default this only trains the airplane category, to train other categories in parallel on multiple GPUs please uncomment the respective lines in the script_*.sh files.

View Quality computation

The function to compute view qualites from meshes are in viewpoint_learning/code/DataProcessing.py.

A basic computation can be done via:

python DataProcessing.py --generate_views --f DATA_DIR

which computes Visibility Ratio, Viewpoint Entropy, Viewpoint Kullback-Leibler Distance and Viewpoint Mutual Information.